1 /*
   2  * Copyright (c) 2003, 2018, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/macroAssembler.hpp"
  27 #include "interpreter/bytecodeHistogram.hpp"
  28 #include "interpreter/interp_masm.hpp"
  29 #include "interpreter/interpreter.hpp"
  30 #include "interpreter/interpreterRuntime.hpp"
  31 #include "interpreter/templateInterpreterGenerator.hpp"
  32 #include "interpreter/templateTable.hpp"
  33 #include "oops/arrayOop.hpp"
  34 #include "oops/methodData.hpp"
  35 #include "oops/method.hpp"
  36 #include "oops/oop.inline.hpp"
  37 #include "prims/jvmtiExport.hpp"
  38 #include "prims/jvmtiThreadState.hpp"
  39 #include "runtime/arguments.hpp"
  40 #include "runtime/deoptimization.hpp"
  41 #include "runtime/frame.inline.hpp"
  42 #include "runtime/sharedRuntime.hpp"
  43 #include "runtime/stubRoutines.hpp"
  44 #include "runtime/synchronizer.hpp"
  45 #include "runtime/timer.hpp"
  46 #include "runtime/vframeArray.hpp"
  47 #include "utilities/debug.hpp"
  48 #include "utilities/macros.hpp"
  49 
  50 #define __ _masm->
  51 
  52 // Size of interpreter code.  Increase if too small.  Interpreter will
  53 // fail with a guarantee ("not enough space for interpreter generation");
  54 // if too small.
  55 // Run with +PrintInterpreter to get the VM to print out the size.
  56 // Max size with JVMTI
  57 #ifdef AMD64
  58 int TemplateInterpreter::InterpreterCodeSize = JVMCI_ONLY(268) NOT_JVMCI(256) * 1024;
  59 #else
  60 int TemplateInterpreter::InterpreterCodeSize = 224 * 1024;
  61 #endif // AMD64
  62 
  63 // Global Register Names
  64 static const Register rbcp     = LP64_ONLY(r13) NOT_LP64(rsi);
  65 static const Register rlocals  = LP64_ONLY(r14) NOT_LP64(rdi);
  66 
  67 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
  68 const int bcp_offset    = frame::interpreter_frame_bcp_offset    * wordSize;
  69 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
  70 
  71 
  72 //-----------------------------------------------------------------------------
  73 
  74 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
  75   address entry = __ pc();
  76 
  77 #ifdef ASSERT
  78   {
  79     Label L;
  80     __ lea(rax, Address(rbp,
  81                         frame::interpreter_frame_monitor_block_top_offset *
  82                         wordSize));
  83     __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
  84                          // grows negative)
  85     __ jcc(Assembler::aboveEqual, L); // check if frame is complete
  86     __ stop ("interpreter frame not set up");
  87     __ bind(L);
  88   }
  89 #endif // ASSERT
  90   // Restore bcp under the assumption that the current frame is still
  91   // interpreted
  92   __ restore_bcp();
  93 
  94   // expression stack must be empty before entering the VM if an
  95   // exception happened
  96   __ empty_expression_stack();
  97   // throw exception
  98   __ call_VM(noreg,
  99              CAST_FROM_FN_PTR(address,
 100                               InterpreterRuntime::throw_StackOverflowError));
 101   return entry;
 102 }
 103 
 104 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
 105         const char* name) {
 106   address entry = __ pc();
 107   // expression stack must be empty before entering the VM if an
 108   // exception happened
 109   __ empty_expression_stack();
 110   // setup parameters
 111   // ??? convention: expect aberrant index in register ebx
 112   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 113   __ lea(rarg, ExternalAddress((address)name));
 114   __ call_VM(noreg,
 115              CAST_FROM_FN_PTR(address,
 116                               InterpreterRuntime::
 117                               throw_ArrayIndexOutOfBoundsException),
 118              rarg, rbx);
 119   return entry;
 120 }
 121 
 122 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 123   address entry = __ pc();
 124 
 125   // object is at TOS
 126   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 127   __ pop(rarg);
 128 
 129   // expression stack must be empty before entering the VM if an
 130   // exception happened
 131   __ empty_expression_stack();
 132 
 133   __ call_VM(noreg,
 134              CAST_FROM_FN_PTR(address,
 135                               InterpreterRuntime::
 136                               throw_ClassCastException),
 137              rarg);
 138   return entry;
 139 }
 140 
 141 address TemplateInterpreterGenerator::generate_exception_handler_common(
 142         const char* name, const char* message, bool pass_oop) {
 143   assert(!pass_oop || message == NULL, "either oop or message but not both");
 144   address entry = __ pc();
 145 
 146   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 147   Register rarg2 = NOT_LP64(rbx) LP64_ONLY(c_rarg2);
 148 
 149   if (pass_oop) {
 150     // object is at TOS
 151     __ pop(rarg2);
 152   }
 153   // expression stack must be empty before entering the VM if an
 154   // exception happened
 155   __ empty_expression_stack();
 156   // setup parameters
 157   __ lea(rarg, ExternalAddress((address)name));
 158   if (pass_oop) {
 159     __ call_VM(rax, CAST_FROM_FN_PTR(address,
 160                                      InterpreterRuntime::
 161                                      create_klass_exception),
 162                rarg, rarg2);
 163   } else {
 164     __ lea(rarg2, ExternalAddress((address)message));
 165     __ call_VM(rax,
 166                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
 167                rarg, rarg2);
 168   }
 169   // throw exception
 170   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
 171   return entry;
 172 }
 173 
 174 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
 175   address entry = __ pc();
 176 
 177 #ifndef _LP64
 178 #ifdef COMPILER2
 179   // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
 180   if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
 181     for (int i = 1; i < 8; i++) {
 182         __ ffree(i);
 183     }
 184   } else if (UseSSE < 2) {
 185     __ empty_FPU_stack();
 186   }
 187 #endif // COMPILER2
 188   if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
 189     __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
 190   } else {
 191     __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
 192   }
 193 
 194   if (state == ftos) {
 195     __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_return_entry_for in interpreter");
 196   } else if (state == dtos) {
 197     __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_return_entry_for in interpreter");
 198   }
 199 #endif // _LP64
 200 
 201   // Restore stack bottom in case i2c adjusted stack
 202   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
 203   // and NULL it as marker that esp is now tos until next java call
 204   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 205 
 206   __ restore_bcp();
 207   __ restore_locals();
 208 
 209   if (state == atos) {
 210     Register mdp = rbx;
 211     Register tmp = rcx;
 212     __ profile_return_type(mdp, rax, tmp);
 213   }
 214 
 215   const Register cache = rbx;
 216   const Register index = rcx;
 217   __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
 218 
 219   const Register flags = cache;
 220   __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
 221   __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
 222   __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
 223 
 224    const Register java_thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
 225    if (JvmtiExport::can_pop_frame()) {
 226      NOT_LP64(__ get_thread(java_thread));
 227      __ check_and_handle_popframe(java_thread);
 228    }
 229    if (JvmtiExport::can_force_early_return()) {
 230      NOT_LP64(__ get_thread(java_thread));
 231      __ check_and_handle_earlyret(java_thread);
 232    }
 233 
 234   __ dispatch_next(state, step);
 235 
 236   return entry;
 237 }
 238 
 239 
 240 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) {
 241   address entry = __ pc();
 242 
 243 #ifndef _LP64
 244   if (state == ftos) {
 245     __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_deopt_entry_for in interpreter");
 246   } else if (state == dtos) {
 247     __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_deopt_entry_for in interpreter");
 248   }
 249 #endif // _LP64
 250 
 251   // NULL last_sp until next java call
 252   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 253   __ restore_bcp();
 254   __ restore_locals();
 255   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
 256   NOT_LP64(__ get_thread(thread));
 257 #if INCLUDE_JVMCI
 258   // Check if we need to take lock at entry of synchronized method.  This can
 259   // only occur on method entry so emit it only for vtos with step 0.
 260   if ((EnableJVMCI || UseAOT) && state == vtos && step == 0) {
 261     Label L;
 262     __ cmpb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
 263     __ jcc(Assembler::zero, L);
 264     // Clear flag.
 265     __ movb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
 266     // Satisfy calling convention for lock_method().
 267     __ get_method(rbx);
 268     // Take lock.
 269     lock_method();
 270     __ bind(L);
 271   } else {
 272 #ifdef ASSERT
 273     if (EnableJVMCI) {
 274       Label L;
 275       __ cmpb(Address(r15_thread, JavaThread::pending_monitorenter_offset()), 0);
 276       __ jccb(Assembler::zero, L);
 277       __ stop("unexpected pending monitor in deopt entry");
 278       __ bind(L);
 279     }
 280 #endif
 281   }
 282 #endif
 283   // handle exceptions
 284   {
 285     Label L;
 286     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
 287     __ jcc(Assembler::zero, L);
 288     __ call_VM(noreg,
 289                CAST_FROM_FN_PTR(address,
 290                                 InterpreterRuntime::throw_pending_exception));
 291     __ should_not_reach_here();
 292     __ bind(L);
 293   }
 294   if (continuation == NULL) {
 295     __ dispatch_next(state, step);
 296   } else {
 297     __ jump_to_entry(continuation);
 298   }
 299   return entry;
 300 }
 301 
 302 address TemplateInterpreterGenerator::generate_result_handler_for(
 303         BasicType type) {
 304   address entry = __ pc();
 305   switch (type) {
 306   case T_BOOLEAN: __ c2bool(rax);            break;
 307 #ifndef _LP64
 308   case T_CHAR   : __ andptr(rax, 0xFFFF);    break;
 309 #else
 310   case T_CHAR   : __ movzwl(rax, rax);       break;
 311 #endif // _LP64
 312   case T_BYTE   : __ sign_extend_byte(rax);  break;
 313   case T_SHORT  : __ sign_extend_short(rax); break;
 314   case T_INT    : /* nothing to do */        break;
 315   case T_LONG   : /* nothing to do */        break;
 316   case T_VOID   : /* nothing to do */        break;
 317 #ifndef _LP64
 318   case T_DOUBLE :
 319   case T_FLOAT  :
 320     { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
 321       __ pop(t);                            // remove return address first
 322       // Must return a result for interpreter or compiler. In SSE
 323       // mode, results are returned in xmm0 and the FPU stack must
 324       // be empty.
 325       if (type == T_FLOAT && UseSSE >= 1) {
 326         // Load ST0
 327         __ fld_d(Address(rsp, 0));
 328         // Store as float and empty fpu stack
 329         __ fstp_s(Address(rsp, 0));
 330         // and reload
 331         __ movflt(xmm0, Address(rsp, 0));
 332       } else if (type == T_DOUBLE && UseSSE >= 2 ) {
 333         __ movdbl(xmm0, Address(rsp, 0));
 334       } else {
 335         // restore ST0
 336         __ fld_d(Address(rsp, 0));
 337       }
 338       // and pop the temp
 339       __ addptr(rsp, 2 * wordSize);
 340       __ push(t);                           // restore return address
 341     }
 342     break;
 343 #else
 344   case T_FLOAT  : /* nothing to do */        break;
 345   case T_DOUBLE : /* nothing to do */        break;
 346 #endif // _LP64
 347 
 348   case T_OBJECT :
 349     // retrieve result from frame
 350     __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
 351     // and verify it
 352     __ verify_oop(rax);
 353     break;
 354   default       : ShouldNotReachHere();
 355   }
 356   __ ret(0);                                   // return from result handler
 357   return entry;
 358 }
 359 
 360 address TemplateInterpreterGenerator::generate_safept_entry_for(
 361         TosState state,
 362         address runtime_entry) {
 363   address entry = __ pc();
 364   __ push(state);
 365   __ call_VM(noreg, runtime_entry);
 366   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 367   return entry;
 368 }
 369 
 370 
 371 
 372 // Helpers for commoning out cases in the various type of method entries.
 373 //
 374 
 375 
 376 // increment invocation count & check for overflow
 377 //
 378 // Note: checking for negative value instead of overflow
 379 //       so we have a 'sticky' overflow test
 380 //
 381 // rbx: method
 382 // rcx: invocation counter
 383 //
 384 void TemplateInterpreterGenerator::generate_counter_incr(
 385         Label* overflow,
 386         Label* profile_method,
 387         Label* profile_method_continue) {
 388   Label done;
 389   // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
 390   if (TieredCompilation) {
 391     int increment = InvocationCounter::count_increment;
 392     Label no_mdo;
 393     if (ProfileInterpreter) {
 394       // Are we profiling?
 395       __ movptr(rax, Address(rbx, Method::method_data_offset()));
 396       __ testptr(rax, rax);
 397       __ jccb(Assembler::zero, no_mdo);
 398       // Increment counter in the MDO
 399       const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
 400                                                 in_bytes(InvocationCounter::counter_offset()));
 401       const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
 402       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
 403       __ jmp(done);
 404     }
 405     __ bind(no_mdo);
 406     // Increment counter in MethodCounters
 407     const Address invocation_counter(rax,
 408                   MethodCounters::invocation_counter_offset() +
 409                   InvocationCounter::counter_offset());
 410     __ get_method_counters(rbx, rax, done);
 411     const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
 412     __ increment_mask_and_jump(invocation_counter, increment, mask, rcx,
 413                                false, Assembler::zero, overflow);
 414     __ bind(done);
 415   } else { // not TieredCompilation
 416     const Address backedge_counter(rax,
 417                   MethodCounters::backedge_counter_offset() +
 418                   InvocationCounter::counter_offset());
 419     const Address invocation_counter(rax,
 420                   MethodCounters::invocation_counter_offset() +
 421                   InvocationCounter::counter_offset());
 422 
 423     __ get_method_counters(rbx, rax, done);
 424 
 425     if (ProfileInterpreter) {
 426       __ incrementl(Address(rax,
 427               MethodCounters::interpreter_invocation_counter_offset()));
 428     }
 429     // Update standard invocation counters
 430     __ movl(rcx, invocation_counter);
 431     __ incrementl(rcx, InvocationCounter::count_increment);
 432     __ movl(invocation_counter, rcx); // save invocation count
 433 
 434     __ movl(rax, backedge_counter);   // load backedge counter
 435     __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
 436 
 437     __ addl(rcx, rax);                // add both counters
 438 
 439     // profile_method is non-null only for interpreted method so
 440     // profile_method != NULL == !native_call
 441 
 442     if (ProfileInterpreter && profile_method != NULL) {
 443       // Test to see if we should create a method data oop
 444       __ movptr(rax, Address(rbx, Method::method_counters_offset()));
 445       __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
 446       __ jcc(Assembler::less, *profile_method_continue);
 447 
 448       // if no method data exists, go to profile_method
 449       __ test_method_data_pointer(rax, *profile_method);
 450     }
 451 
 452     __ movptr(rax, Address(rbx, Method::method_counters_offset()));
 453     __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_invocation_limit_offset())));
 454     __ jcc(Assembler::aboveEqual, *overflow);
 455     __ bind(done);
 456   }
 457 }
 458 
 459 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
 460 
 461   // Asm interpreter on entry
 462   // r14/rdi - locals
 463   // r13/rsi - bcp
 464   // rbx - method
 465   // rdx - cpool --- DOES NOT APPEAR TO BE TRUE
 466   // rbp - interpreter frame
 467 
 468   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
 469   // Everything as it was on entry
 470   // rdx is not restored. Doesn't appear to really be set.
 471 
 472   // InterpreterRuntime::frequency_counter_overflow takes two
 473   // arguments, the first (thread) is passed by call_VM, the second
 474   // indicates if the counter overflow occurs at a backwards branch
 475   // (NULL bcp).  We pass zero for it.  The call returns the address
 476   // of the verified entry point for the method or NULL if the
 477   // compilation did not complete (either went background or bailed
 478   // out).
 479   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 480   __ movl(rarg, 0);
 481   __ call_VM(noreg,
 482              CAST_FROM_FN_PTR(address,
 483                               InterpreterRuntime::frequency_counter_overflow),
 484              rarg);
 485 
 486   __ movptr(rbx, Address(rbp, method_offset));   // restore Method*
 487   // Preserve invariant that r13/r14 contain bcp/locals of sender frame
 488   // and jump to the interpreted entry.
 489   __ jmp(do_continue, relocInfo::none);
 490 }
 491 
 492 // See if we've got enough room on the stack for locals plus overhead below
 493 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
 494 // without going through the signal handler, i.e., reserved and yellow zones
 495 // will not be made usable. The shadow zone must suffice to handle the
 496 // overflow.
 497 // The expression stack grows down incrementally, so the normal guard
 498 // page mechanism will work for that.
 499 //
 500 // NOTE: Since the additional locals are also always pushed (wasn't
 501 // obvious in generate_fixed_frame) so the guard should work for them
 502 // too.
 503 //
 504 // Args:
 505 //      rdx: number of additional locals this frame needs (what we must check)
 506 //      rbx: Method*
 507 //
 508 // Kills:
 509 //      rax
 510 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
 511 
 512   // monitor entry size: see picture of stack in frame_x86.hpp
 513   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 514 
 515   // total overhead size: entry_size + (saved rbp through expr stack
 516   // bottom).  be sure to change this if you add/subtract anything
 517   // to/from the overhead area
 518   const int overhead_size =
 519     -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
 520 
 521   const int page_size = os::vm_page_size();
 522 
 523   Label after_frame_check;
 524 
 525   // see if the frame is greater than one page in size. If so,
 526   // then we need to verify there is enough stack space remaining
 527   // for the additional locals.
 528   __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
 529   __ jcc(Assembler::belowEqual, after_frame_check);
 530 
 531   // compute rsp as if this were going to be the last frame on
 532   // the stack before the red zone
 533 
 534   Label after_frame_check_pop;
 535   const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
 536 #ifndef _LP64
 537   __ push(thread);
 538   __ get_thread(thread);
 539 #endif
 540 
 541   const Address stack_limit(thread, JavaThread::stack_overflow_limit_offset());
 542 
 543   // locals + overhead, in bytes
 544   __ mov(rax, rdx);
 545   __ shlptr(rax, Interpreter::logStackElementSize); // Convert parameter count to bytes.
 546   __ addptr(rax, overhead_size);
 547 
 548 #ifdef ASSERT
 549   Label limit_okay;
 550   // Verify that thread stack overflow limit is non-zero.
 551   __ cmpptr(stack_limit, (int32_t)NULL_WORD);
 552   __ jcc(Assembler::notEqual, limit_okay);
 553   __ stop("stack overflow limit is zero");
 554   __ bind(limit_okay);
 555 #endif
 556 
 557   // Add locals/frame size to stack limit.
 558   __ addptr(rax, stack_limit);
 559 
 560   // Check against the current stack bottom.
 561   __ cmpptr(rsp, rax);
 562 
 563   __ jcc(Assembler::above, after_frame_check_pop);
 564   NOT_LP64(__ pop(rsi));  // get saved bcp
 565 
 566   // Restore sender's sp as SP. This is necessary if the sender's
 567   // frame is an extended compiled frame (see gen_c2i_adapter())
 568   // and safer anyway in case of JSR292 adaptations.
 569 
 570   __ pop(rax); // return address must be moved if SP is changed
 571   __ mov(rsp, rbcp);
 572   __ push(rax);
 573 
 574   // Note: the restored frame is not necessarily interpreted.
 575   // Use the shared runtime version of the StackOverflowError.
 576   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
 577   __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
 578   // all done with frame size check
 579   __ bind(after_frame_check_pop);
 580   NOT_LP64(__ pop(rsi));
 581 
 582   // all done with frame size check
 583   __ bind(after_frame_check);
 584 }
 585 
 586 // Allocate monitor and lock method (asm interpreter)
 587 //
 588 // Args:
 589 //      rbx: Method*
 590 //      r14/rdi: locals
 591 //
 592 // Kills:
 593 //      rax
 594 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
 595 //      rscratch1, rscratch2 (scratch regs)
 596 void TemplateInterpreterGenerator::lock_method() {
 597   // synchronize method
 598   const Address access_flags(rbx, Method::access_flags_offset());
 599   const Address monitor_block_top(
 600         rbp,
 601         frame::interpreter_frame_monitor_block_top_offset * wordSize);
 602   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 603 
 604 #ifdef ASSERT
 605   {
 606     Label L;
 607     __ movl(rax, access_flags);
 608     __ testl(rax, JVM_ACC_SYNCHRONIZED);
 609     __ jcc(Assembler::notZero, L);
 610     __ stop("method doesn't need synchronization");
 611     __ bind(L);
 612   }
 613 #endif // ASSERT
 614 
 615   // get synchronization object
 616   {
 617     Label done;
 618     __ movl(rax, access_flags);
 619     __ testl(rax, JVM_ACC_STATIC);
 620     // get receiver (assume this is frequent case)
 621     __ movptr(rax, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
 622     __ jcc(Assembler::zero, done);
 623     __ load_mirror(rax, rbx);
 624 
 625 #ifdef ASSERT
 626     {
 627       Label L;
 628       __ testptr(rax, rax);
 629       __ jcc(Assembler::notZero, L);
 630       __ stop("synchronization object is NULL");
 631       __ bind(L);
 632     }
 633 #endif // ASSERT
 634 
 635     __ bind(done);
 636   }
 637 
 638   // add space for monitor & lock
 639   __ subptr(rsp, entry_size); // add space for a monitor entry
 640   __ movptr(monitor_block_top, rsp);  // set new monitor block top
 641   // store object
 642   __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
 643   const Register lockreg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
 644   __ movptr(lockreg, rsp); // object address
 645   __ lock_object(lockreg);
 646 }
 647 
 648 // Generate a fixed interpreter frame. This is identical setup for
 649 // interpreted methods and for native methods hence the shared code.
 650 //
 651 // Args:
 652 //      rax: return address
 653 //      rbx: Method*
 654 //      r14/rdi: pointer to locals
 655 //      r13/rsi: sender sp
 656 //      rdx: cp cache
 657 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 658   // initialize fixed part of activation frame
 659   __ push(rax);        // save return address
 660   __ enter();          // save old & set new rbp
 661   __ push(rbcp);        // set sender sp
 662   __ push((int)NULL_WORD); // leave last_sp as null
 663   __ movptr(rbcp, Address(rbx, Method::const_offset()));      // get ConstMethod*
 664   __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
 665   __ push(rbx);        // save Method*
 666   // Get mirror and store it in the frame as GC root for this Method*
 667   __ load_mirror(rdx, rbx);
 668   __ push(rdx);
 669   if (ProfileInterpreter) {
 670     Label method_data_continue;
 671     __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
 672     __ testptr(rdx, rdx);
 673     __ jcc(Assembler::zero, method_data_continue);
 674     __ addptr(rdx, in_bytes(MethodData::data_offset()));
 675     __ bind(method_data_continue);
 676     __ push(rdx);      // set the mdp (method data pointer)
 677   } else {
 678     __ push(0);
 679   }
 680 
 681   __ movptr(rdx, Address(rbx, Method::const_offset()));
 682   __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
 683   __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
 684   __ push(rdx); // set constant pool cache
 685   __ push(rlocals); // set locals pointer
 686   if (native_call) {
 687     __ push(0); // no bcp
 688   } else {
 689     __ push(rbcp); // set bcp
 690   }
 691   __ push(0); // reserve word for pointer to expression stack bottom
 692   __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
 693 }
 694 
 695 // End of helpers
 696 
 697 // Method entry for java.lang.ref.Reference.get.
 698 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
 699 #if INCLUDE_ALL_GCS
 700   // Code: _aload_0, _getfield, _areturn
 701   // parameter size = 1
 702   //
 703   // The code that gets generated by this routine is split into 2 parts:
 704   //    1. The "intrinsified" code for G1 (or any SATB based GC),
 705   //    2. The slow path - which is an expansion of the regular method entry.
 706   //
 707   // Notes:-
 708   // * In the G1 code we do not check whether we need to block for
 709   //   a safepoint. If G1 is enabled then we must execute the specialized
 710   //   code for Reference.get (except when the Reference object is null)
 711   //   so that we can log the value in the referent field with an SATB
 712   //   update buffer.
 713   //   If the code for the getfield template is modified so that the
 714   //   G1 pre-barrier code is executed when the current method is
 715   //   Reference.get() then going through the normal method entry
 716   //   will be fine.
 717   // * The G1 code can, however, check the receiver object (the instance
 718   //   of java.lang.Reference) and jump to the slow path if null. If the
 719   //   Reference object is null then we obviously cannot fetch the referent
 720   //   and so we don't need to call the G1 pre-barrier. Thus we can use the
 721   //   regular method entry code to generate the NPE.
 722   //
 723   // rbx: Method*
 724 
 725   // r13: senderSP must preserve for slow path, set SP to it on fast path
 726 
 727   address entry = __ pc();
 728 
 729   const int referent_offset = java_lang_ref_Reference::referent_offset;
 730   guarantee(referent_offset > 0, "referent offset not initialized");
 731 
 732   if (UseG1GC) {
 733     Label slow_path;
 734     // rbx: method
 735 
 736     // Check if local 0 != NULL
 737     // If the receiver is null then it is OK to jump to the slow path.
 738     __ movptr(rax, Address(rsp, wordSize));
 739 
 740     __ testptr(rax, rax);
 741     __ jcc(Assembler::zero, slow_path);
 742 
 743     // rax: local 0
 744     // rbx: method (but can be used as scratch now)
 745     // rdx: scratch
 746     // rdi: scratch
 747 
 748     // Preserve the sender sp in case the pre-barrier
 749     // calls the runtime
 750     NOT_LP64(__ push(rsi));
 751 
 752     // Generate the G1 pre-barrier code to log the value of
 753     // the referent field in an SATB buffer.
 754 
 755     // Load the value of the referent field.
 756     const Address field_address(rax, referent_offset);
 757     __ load_heap_oop(rax, field_address);
 758 
 759     const Register sender_sp = NOT_LP64(rsi) LP64_ONLY(r13);
 760     const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
 761     NOT_LP64(__ get_thread(thread));
 762 
 763     // Generate the G1 pre-barrier code to log the value of
 764     // the referent field in an SATB buffer.
 765     __ g1_write_barrier_pre(noreg /* obj */,
 766                             rax /* pre_val */,
 767                             thread /* thread */,
 768                             rbx /* tmp */,
 769                             true /* tosca_live */,
 770                             true /* expand_call */);
 771 
 772     // _areturn
 773     NOT_LP64(__ pop(rsi));      // get sender sp
 774     __ pop(rdi);                // get return address
 775     __ mov(rsp, sender_sp);     // set sp to sender sp
 776     __ jmp(rdi);
 777     __ ret(0);
 778 
 779     // generate a vanilla interpreter entry as the slow path
 780     __ bind(slow_path);
 781     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
 782     return entry;
 783   }
 784 #endif // INCLUDE_ALL_GCS
 785 
 786   // If G1 is not enabled then attempt to go through the accessor entry point
 787   // Reference.get is an accessor
 788   return NULL;
 789 }
 790 
 791 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
 792   // Quick & dirty stack overflow checking: bang the stack & handle trap.
 793   // Note that we do the banging after the frame is setup, since the exception
 794   // handling code expects to find a valid interpreter frame on the stack.
 795   // Doing the banging earlier fails if the caller frame is not an interpreter
 796   // frame.
 797   // (Also, the exception throwing code expects to unlock any synchronized
 798   // method receiever, so do the banging after locking the receiver.)
 799 
 800   // Bang each page in the shadow zone. We can't assume it's been done for
 801   // an interpreter frame with greater than a page of locals, so each page
 802   // needs to be checked.  Only true for non-native.
 803   if (UseStackBanging) {
 804     const int page_size = os::vm_page_size();
 805     const int n_shadow_pages = ((int)JavaThread::stack_shadow_zone_size()) / page_size;
 806     const int start_page = native_call ? n_shadow_pages : 1;
 807     for (int pages = start_page; pages <= n_shadow_pages; pages++) {
 808       __ bang_stack_with_offset(pages*page_size);
 809     }
 810   }
 811 }
 812 
 813 // Interpreter stub for calling a native method. (asm interpreter)
 814 // This sets up a somewhat different looking stack for calling the
 815 // native method than the typical interpreter frame setup.
 816 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
 817   // determine code generation flags
 818   bool inc_counter  = UseCompiler || CountCompiledCalls || LogTouchedMethods;
 819 
 820   // rbx: Method*
 821   // rbcp: sender sp
 822 
 823   address entry_point = __ pc();
 824 
 825   const Address constMethod       (rbx, Method::const_offset());
 826   const Address access_flags      (rbx, Method::access_flags_offset());
 827   const Address size_of_parameters(rcx, ConstMethod::
 828                                         size_of_parameters_offset());
 829 
 830 
 831   // get parameter size (always needed)
 832   __ movptr(rcx, constMethod);
 833   __ load_unsigned_short(rcx, size_of_parameters);
 834 
 835   // native calls don't need the stack size check since they have no
 836   // expression stack and the arguments are already on the stack and
 837   // we only add a handful of words to the stack
 838 
 839   // rbx: Method*
 840   // rcx: size of parameters
 841   // rbcp: sender sp
 842   __ pop(rax);                                       // get return address
 843 
 844   // for natives the size of locals is zero
 845 
 846   // compute beginning of parameters
 847   __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
 848 
 849   // add 2 zero-initialized slots for native calls
 850   // initialize result_handler slot
 851   __ push((int) NULL_WORD);
 852   // slot for oop temp
 853   // (static native method holder mirror/jni oop result)
 854   __ push((int) NULL_WORD);
 855 
 856   // initialize fixed part of activation frame
 857   generate_fixed_frame(true);
 858 
 859   // make sure method is native & not abstract
 860 #ifdef ASSERT
 861   __ movl(rax, access_flags);
 862   {
 863     Label L;
 864     __ testl(rax, JVM_ACC_NATIVE);
 865     __ jcc(Assembler::notZero, L);
 866     __ stop("tried to execute non-native method as native");
 867     __ bind(L);
 868   }
 869   {
 870     Label L;
 871     __ testl(rax, JVM_ACC_ABSTRACT);
 872     __ jcc(Assembler::zero, L);
 873     __ stop("tried to execute abstract method in interpreter");
 874     __ bind(L);
 875   }
 876 #endif
 877 
 878   // Since at this point in the method invocation the exception handler
 879   // would try to exit the monitor of synchronized methods which hasn't
 880   // been entered yet, we set the thread local variable
 881   // _do_not_unlock_if_synchronized to true. The remove_activation will
 882   // check this flag.
 883 
 884   const Register thread1 = NOT_LP64(rax) LP64_ONLY(r15_thread);
 885   NOT_LP64(__ get_thread(thread1));
 886   const Address do_not_unlock_if_synchronized(thread1,
 887         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 888   __ movbool(do_not_unlock_if_synchronized, true);
 889 
 890   // increment invocation count & check for overflow
 891   Label invocation_counter_overflow;
 892   if (inc_counter) {
 893     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
 894   }
 895 
 896   Label continue_after_compile;
 897   __ bind(continue_after_compile);
 898 
 899   bang_stack_shadow_pages(true);
 900 
 901   // reset the _do_not_unlock_if_synchronized flag
 902   NOT_LP64(__ get_thread(thread1));
 903   __ movbool(do_not_unlock_if_synchronized, false);
 904 
 905   // check for synchronized methods
 906   // Must happen AFTER invocation_counter check and stack overflow check,
 907   // so method is not locked if overflows.
 908   if (synchronized) {
 909     lock_method();
 910   } else {
 911     // no synchronization necessary
 912 #ifdef ASSERT
 913     {
 914       Label L;
 915       __ movl(rax, access_flags);
 916       __ testl(rax, JVM_ACC_SYNCHRONIZED);
 917       __ jcc(Assembler::zero, L);
 918       __ stop("method needs synchronization");
 919       __ bind(L);
 920     }
 921 #endif
 922   }
 923 
 924   // start execution
 925 #ifdef ASSERT
 926   {
 927     Label L;
 928     const Address monitor_block_top(rbp,
 929                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
 930     __ movptr(rax, monitor_block_top);
 931     __ cmpptr(rax, rsp);
 932     __ jcc(Assembler::equal, L);
 933     __ stop("broken stack frame setup in interpreter");
 934     __ bind(L);
 935   }
 936 #endif
 937 
 938   // jvmti support
 939   __ notify_method_entry();
 940 
 941   // work registers
 942   const Register method = rbx;
 943   const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
 944   const Register t      = NOT_LP64(rcx) LP64_ONLY(r11);
 945 
 946   // allocate space for parameters
 947   __ get_method(method);
 948   __ movptr(t, Address(method, Method::const_offset()));
 949   __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
 950 
 951 #ifndef _LP64
 952   __ shlptr(t, Interpreter::logStackElementSize); // Convert parameter count to bytes.
 953   __ addptr(t, 2*wordSize);     // allocate two more slots for JNIEnv and possible mirror
 954   __ subptr(rsp, t);
 955   __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
 956 #else
 957   __ shll(t, Interpreter::logStackElementSize);
 958 
 959   __ subptr(rsp, t);
 960   __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
 961   __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
 962 #endif // _LP64
 963 
 964   // get signature handler
 965   {
 966     Label L;
 967     __ movptr(t, Address(method, Method::signature_handler_offset()));
 968     __ testptr(t, t);
 969     __ jcc(Assembler::notZero, L);
 970     __ call_VM(noreg,
 971                CAST_FROM_FN_PTR(address,
 972                                 InterpreterRuntime::prepare_native_call),
 973                method);
 974     __ get_method(method);
 975     __ movptr(t, Address(method, Method::signature_handler_offset()));
 976     __ bind(L);
 977   }
 978 
 979   // call signature handler
 980   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
 981          "adjust this code");
 982   assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
 983          "adjust this code");
 984   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == NOT_LP64(t) LP64_ONLY(rscratch1),
 985          "adjust this code");
 986 
 987   // The generated handlers do not touch RBX (the method oop).
 988   // However, large signatures cannot be cached and are generated
 989   // each time here.  The slow-path generator can do a GC on return,
 990   // so we must reload it after the call.
 991   __ call(t);
 992   __ get_method(method);        // slow path can do a GC, reload RBX
 993 
 994 
 995   // result handler is in rax
 996   // set result handler
 997   __ movptr(Address(rbp,
 998                     (frame::interpreter_frame_result_handler_offset) * wordSize),
 999             rax);
1000 
1001   // pass mirror handle if static call
1002   {
1003     Label L;
1004     __ movl(t, Address(method, Method::access_flags_offset()));
1005     __ testl(t, JVM_ACC_STATIC);
1006     __ jcc(Assembler::zero, L);
1007     // get mirror
1008     __ load_mirror(t, method);
1009     // copy mirror into activation frame
1010     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
1011             t);
1012     // pass handle to mirror
1013 #ifndef _LP64
1014     __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1015     __ movptr(Address(rsp, wordSize), t);
1016 #else
1017     __ lea(c_rarg1,
1018            Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1019 #endif // _LP64
1020     __ bind(L);
1021   }
1022 
1023   // get native function entry point
1024   {
1025     Label L;
1026     __ movptr(rax, Address(method, Method::native_function_offset()));
1027     ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1028     __ cmpptr(rax, unsatisfied.addr());
1029     __ jcc(Assembler::notEqual, L);
1030     __ call_VM(noreg,
1031                CAST_FROM_FN_PTR(address,
1032                                 InterpreterRuntime::prepare_native_call),
1033                method);
1034     __ get_method(method);
1035     __ movptr(rax, Address(method, Method::native_function_offset()));
1036     __ bind(L);
1037   }
1038 
1039   // pass JNIEnv
1040 #ifndef _LP64
1041    __ get_thread(thread);
1042    __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1043    __ movptr(Address(rsp, 0), t);
1044 
1045    // set_last_Java_frame_before_call
1046    // It is enough that the pc()
1047    // points into the right code segment. It does not have to be the correct return pc.
1048    __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1049 #else
1050    __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1051 
1052    // It is enough that the pc() points into the right code
1053    // segment. It does not have to be the correct return pc.
1054    __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1055 #endif // _LP64
1056 
1057   // change thread state
1058 #ifdef ASSERT
1059   {
1060     Label L;
1061     __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1062     __ cmpl(t, _thread_in_Java);
1063     __ jcc(Assembler::equal, L);
1064     __ stop("Wrong thread state in native stub");
1065     __ bind(L);
1066   }
1067 #endif
1068 
1069   // Change state to native
1070 
1071   __ movl(Address(thread, JavaThread::thread_state_offset()),
1072           _thread_in_native);
1073 
1074   // Call the native method.
1075   __ call(rax);
1076   // 32: result potentially in rdx:rax or ST0
1077   // 64: result potentially in rax or xmm0
1078 
1079   // Verify or restore cpu control state after JNI call
1080   __ restore_cpu_control_state_after_jni();
1081 
1082   // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1083   // in order to extract the result of a method call. If the order of these
1084   // pushes change or anything else is added to the stack then the code in
1085   // interpreter_frame_result must also change.
1086 
1087 #ifndef _LP64
1088   // save potential result in ST(0) & rdx:rax
1089   // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1090   // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1091   // It is safe to do this push because state is _thread_in_native and return address will be found
1092   // via _last_native_pc and not via _last_jave_sp
1093 
1094   // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1095   // If the order changes or anything else is added to the stack the code in
1096   // interpreter_frame_result will have to be changed.
1097 
1098   { Label L;
1099     Label push_double;
1100     ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1101     ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1102     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1103               float_handler.addr());
1104     __ jcc(Assembler::equal, push_double);
1105     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1106               double_handler.addr());
1107     __ jcc(Assembler::notEqual, L);
1108     __ bind(push_double);
1109     __ push_d(); // FP values are returned using the FPU, so push FPU contents (even if UseSSE > 0).
1110     __ bind(L);
1111   }
1112 #else
1113   __ push(dtos);
1114 #endif // _LP64
1115 
1116   __ push(ltos);
1117 
1118   // change thread state
1119   NOT_LP64(__ get_thread(thread));
1120   __ movl(Address(thread, JavaThread::thread_state_offset()),
1121           _thread_in_native_trans);
1122 
1123   if (os::is_MP()) {
1124     if (UseMembar) {
1125       // Force this write out before the read below
1126       __ membar(Assembler::Membar_mask_bits(
1127            Assembler::LoadLoad | Assembler::LoadStore |
1128            Assembler::StoreLoad | Assembler::StoreStore));
1129     } else {
1130       // Write serialization page so VM thread can do a pseudo remote membar.
1131       // We use the current thread pointer to calculate a thread specific
1132       // offset to write to within the page. This minimizes bus traffic
1133       // due to cache line collision.
1134       __ serialize_memory(thread, rcx);
1135     }
1136   }
1137 
1138 #ifndef _LP64
1139   if (AlwaysRestoreFPU) {
1140     //  Make sure the control word is correct.
1141     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1142   }
1143 #endif // _LP64
1144 
1145   // check for safepoint operation in progress and/or pending suspend requests
1146   {
1147     Label Continue;
1148     Label slow_path;
1149 
1150 #ifndef _LP64
1151     __ safepoint_poll(slow_path, thread, noreg);
1152 #else
1153     __ safepoint_poll(slow_path, r15_thread, rscratch1);
1154 #endif
1155 
1156     __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1157     __ jcc(Assembler::equal, Continue);
1158     __ bind(slow_path);
1159 
1160     // Don't use call_VM as it will see a possible pending exception
1161     // and forward it and never return here preventing us from
1162     // clearing _last_native_pc down below.  Also can't use
1163     // call_VM_leaf either as it will check to see if r13 & r14 are
1164     // preserved and correspond to the bcp/locals pointers. So we do a
1165     // runtime call by hand.
1166     //
1167 #ifndef _LP64
1168     __ push(thread);
1169     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1170                                             JavaThread::check_special_condition_for_native_trans)));
1171     __ increment(rsp, wordSize);
1172     __ get_thread(thread);
1173 #else
1174     __ mov(c_rarg0, r15_thread);
1175     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1176     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1177     __ andptr(rsp, -16); // align stack as required by ABI
1178     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1179     __ mov(rsp, r12); // restore sp
1180     __ reinit_heapbase();
1181 #endif // _LP64
1182     __ bind(Continue);
1183   }
1184 
1185   // change thread state
1186   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1187 
1188   // reset_last_Java_frame
1189   __ reset_last_Java_frame(thread, true);
1190 
1191   if (CheckJNICalls) {
1192     // clear_pending_jni_exception_check
1193     __ movptr(Address(thread, JavaThread::pending_jni_exception_check_fn_offset()), NULL_WORD);
1194   }
1195 
1196   // reset handle block
1197   __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1198   __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1199 
1200   // If result is an oop unbox and store it in frame where gc will see it
1201   // and result handler will pick it up
1202 
1203   {
1204     Label no_oop, not_weak, store_result;
1205     __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1206     __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1207     __ jcc(Assembler::notEqual, no_oop);
1208     // retrieve result
1209     __ pop(ltos);
1210     // Unbox oop result, e.g. JNIHandles::resolve value.
1211     __ resolve_jobject(rax /* value */,
1212                        thread /* thread */,
1213                        t /* tmp */);
1214     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1215     // keep stack depth as expected by pushing oop which will eventually be discarded
1216     __ push(ltos);
1217     __ bind(no_oop);
1218   }
1219 
1220 
1221   {
1222     Label no_reguard;
1223     __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()),
1224             JavaThread::stack_guard_yellow_reserved_disabled);
1225     __ jcc(Assembler::notEqual, no_reguard);
1226 
1227     __ pusha(); // XXX only save smashed registers
1228 #ifndef _LP64
1229     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1230     __ popa();
1231 #else
1232     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1233     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1234     __ andptr(rsp, -16); // align stack as required by ABI
1235     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1236     __ mov(rsp, r12); // restore sp
1237     __ popa(); // XXX only restore smashed registers
1238     __ reinit_heapbase();
1239 #endif // _LP64
1240 
1241     __ bind(no_reguard);
1242   }
1243 
1244 
1245   // The method register is junk from after the thread_in_native transition
1246   // until here.  Also can't call_VM until the bcp has been
1247   // restored.  Need bcp for throwing exception below so get it now.
1248   __ get_method(method);
1249 
1250   // restore to have legal interpreter frame, i.e., bci == 0 <=> code_base()
1251   __ movptr(rbcp, Address(method, Method::const_offset()));   // get ConstMethod*
1252   __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset()));    // get codebase
1253 
1254   // handle exceptions (exception handling will handle unlocking!)
1255   {
1256     Label L;
1257     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1258     __ jcc(Assembler::zero, L);
1259     // Note: At some point we may want to unify this with the code
1260     // used in call_VM_base(); i.e., we should use the
1261     // StubRoutines::forward_exception code. For now this doesn't work
1262     // here because the rsp is not correctly set at this point.
1263     __ MacroAssembler::call_VM(noreg,
1264                                CAST_FROM_FN_PTR(address,
1265                                InterpreterRuntime::throw_pending_exception));
1266     __ should_not_reach_here();
1267     __ bind(L);
1268   }
1269 
1270   // do unlocking if necessary
1271   {
1272     Label L;
1273     __ movl(t, Address(method, Method::access_flags_offset()));
1274     __ testl(t, JVM_ACC_SYNCHRONIZED);
1275     __ jcc(Assembler::zero, L);
1276     // the code below should be shared with interpreter macro
1277     // assembler implementation
1278     {
1279       Label unlock;
1280       // BasicObjectLock will be first in list, since this is a
1281       // synchronized method. However, need to check that the object
1282       // has not been unlocked by an explicit monitorexit bytecode.
1283       const Address monitor(rbp,
1284                             (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1285                                        wordSize - (int)sizeof(BasicObjectLock)));
1286 
1287       const Register regmon = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1288 
1289       // monitor expect in c_rarg1 for slow unlock path
1290       __ lea(regmon, monitor); // address of first monitor
1291 
1292       __ movptr(t, Address(regmon, BasicObjectLock::obj_offset_in_bytes()));
1293       __ testptr(t, t);
1294       __ jcc(Assembler::notZero, unlock);
1295 
1296       // Entry already unlocked, need to throw exception
1297       __ MacroAssembler::call_VM(noreg,
1298                                  CAST_FROM_FN_PTR(address,
1299                    InterpreterRuntime::throw_illegal_monitor_state_exception));
1300       __ should_not_reach_here();
1301 
1302       __ bind(unlock);
1303       __ unlock_object(regmon);
1304     }
1305     __ bind(L);
1306   }
1307 
1308   // jvmti support
1309   // Note: This must happen _after_ handling/throwing any exceptions since
1310   //       the exception handler code notifies the runtime of method exits
1311   //       too. If this happens before, method entry/exit notifications are
1312   //       not properly paired (was bug - gri 11/22/99).
1313   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1314 
1315   // restore potential result in edx:eax, call result handler to
1316   // restore potential result in ST0 & handle result
1317 
1318   __ pop(ltos);
1319   LP64_ONLY( __ pop(dtos));
1320 
1321   __ movptr(t, Address(rbp,
1322                        (frame::interpreter_frame_result_handler_offset) * wordSize));
1323   __ call(t);
1324 
1325   // remove activation
1326   __ movptr(t, Address(rbp,
1327                        frame::interpreter_frame_sender_sp_offset *
1328                        wordSize)); // get sender sp
1329   __ leave();                                // remove frame anchor
1330   __ pop(rdi);                               // get return address
1331   __ mov(rsp, t);                            // set sp to sender sp
1332   __ jmp(rdi);
1333 
1334   if (inc_counter) {
1335     // Handle overflow of counter and compile method
1336     __ bind(invocation_counter_overflow);
1337     generate_counter_overflow(continue_after_compile);
1338   }
1339 
1340   return entry_point;
1341 }
1342 
1343 // Abstract method entry
1344 // Attempt to execute abstract method. Throw exception
1345 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
1346 
1347   address entry_point = __ pc();
1348 
1349   // abstract method entry
1350 
1351   //  pop return address, reset last_sp to NULL
1352   __ empty_expression_stack();
1353   __ restore_bcp();      // rsi must be correct for exception handler   (was destroyed)
1354   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
1355 
1356   // throw exception
1357   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
1358   // the call_VM checks for exception, so we should never return here.
1359   __ should_not_reach_here();
1360 
1361   return entry_point;
1362 }
1363 
1364 //
1365 // Generic interpreted method entry to (asm) interpreter
1366 //
1367 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1368   // determine code generation flags
1369   bool inc_counter  = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1370 
1371   // ebx: Method*
1372   // rbcp: sender sp
1373   address entry_point = __ pc();
1374 
1375   const Address constMethod(rbx, Method::const_offset());
1376   const Address access_flags(rbx, Method::access_flags_offset());
1377   const Address size_of_parameters(rdx,
1378                                    ConstMethod::size_of_parameters_offset());
1379   const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1380 
1381 
1382   // get parameter size (always needed)
1383   __ movptr(rdx, constMethod);
1384   __ load_unsigned_short(rcx, size_of_parameters);
1385 
1386   // rbx: Method*
1387   // rcx: size of parameters
1388   // rbcp: sender_sp (could differ from sp+wordSize if we were called via c2i )
1389 
1390   __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1391   __ subl(rdx, rcx); // rdx = no. of additional locals
1392 
1393   // YYY
1394 //   __ incrementl(rdx);
1395 //   __ andl(rdx, -2);
1396 
1397   // see if we've got enough room on the stack for locals plus overhead.
1398   generate_stack_overflow_check();
1399 
1400   // get return address
1401   __ pop(rax);
1402 
1403   // compute beginning of parameters
1404   __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1405 
1406   // rdx - # of additional locals
1407   // allocate space for locals
1408   // explicitly initialize locals
1409   {
1410     Label exit, loop;
1411     __ testl(rdx, rdx);
1412     __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1413     __ bind(loop);
1414     __ push((int) NULL_WORD); // initialize local variables
1415     __ decrementl(rdx); // until everything initialized
1416     __ jcc(Assembler::greater, loop);
1417     __ bind(exit);
1418   }
1419 
1420   // initialize fixed part of activation frame
1421   generate_fixed_frame(false);
1422 
1423   // make sure method is not native & not abstract
1424 #ifdef ASSERT
1425   __ movl(rax, access_flags);
1426   {
1427     Label L;
1428     __ testl(rax, JVM_ACC_NATIVE);
1429     __ jcc(Assembler::zero, L);
1430     __ stop("tried to execute native method as non-native");
1431     __ bind(L);
1432   }
1433   {
1434     Label L;
1435     __ testl(rax, JVM_ACC_ABSTRACT);
1436     __ jcc(Assembler::zero, L);
1437     __ stop("tried to execute abstract method in interpreter");
1438     __ bind(L);
1439   }
1440 #endif
1441 
1442   // Since at this point in the method invocation the exception
1443   // handler would try to exit the monitor of synchronized methods
1444   // which hasn't been entered yet, we set the thread local variable
1445   // _do_not_unlock_if_synchronized to true. The remove_activation
1446   // will check this flag.
1447 
1448   const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
1449   NOT_LP64(__ get_thread(thread));
1450   const Address do_not_unlock_if_synchronized(thread,
1451         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1452   __ movbool(do_not_unlock_if_synchronized, true);
1453 
1454   __ profile_parameters_type(rax, rcx, rdx);
1455   // increment invocation count & check for overflow
1456   Label invocation_counter_overflow;
1457   Label profile_method;
1458   Label profile_method_continue;
1459   if (inc_counter) {
1460     generate_counter_incr(&invocation_counter_overflow,
1461                           &profile_method,
1462                           &profile_method_continue);
1463     if (ProfileInterpreter) {
1464       __ bind(profile_method_continue);
1465     }
1466   }
1467 
1468   Label continue_after_compile;
1469   __ bind(continue_after_compile);
1470 
1471   // check for synchronized interpreted methods
1472   bang_stack_shadow_pages(false);
1473 
1474   // reset the _do_not_unlock_if_synchronized flag
1475   NOT_LP64(__ get_thread(thread));
1476   __ movbool(do_not_unlock_if_synchronized, false);
1477 
1478   // check for synchronized methods
1479   // Must happen AFTER invocation_counter check and stack overflow check,
1480   // so method is not locked if overflows.
1481   if (synchronized) {
1482     // Allocate monitor and lock method
1483     lock_method();
1484   } else {
1485     // no synchronization necessary
1486 #ifdef ASSERT
1487     {
1488       Label L;
1489       __ movl(rax, access_flags);
1490       __ testl(rax, JVM_ACC_SYNCHRONIZED);
1491       __ jcc(Assembler::zero, L);
1492       __ stop("method needs synchronization");
1493       __ bind(L);
1494     }
1495 #endif
1496   }
1497 
1498   // start execution
1499 #ifdef ASSERT
1500   {
1501     Label L;
1502      const Address monitor_block_top (rbp,
1503                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1504     __ movptr(rax, monitor_block_top);
1505     __ cmpptr(rax, rsp);
1506     __ jcc(Assembler::equal, L);
1507     __ stop("broken stack frame setup in interpreter");
1508     __ bind(L);
1509   }
1510 #endif
1511 
1512   // jvmti support
1513   __ notify_method_entry();
1514 
1515   __ dispatch_next(vtos);
1516 
1517   // invocation counter overflow
1518   if (inc_counter) {
1519     if (ProfileInterpreter) {
1520       // We have decided to profile this method in the interpreter
1521       __ bind(profile_method);
1522       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1523       __ set_method_data_pointer_for_bcp();
1524       __ get_method(rbx);
1525       __ jmp(profile_method_continue);
1526     }
1527     // Handle overflow of counter and compile method
1528     __ bind(invocation_counter_overflow);
1529     generate_counter_overflow(continue_after_compile);
1530   }
1531 
1532   return entry_point;
1533 }
1534 
1535 //-----------------------------------------------------------------------------
1536 // Exceptions
1537 
1538 void TemplateInterpreterGenerator::generate_throw_exception() {
1539   // Entry point in previous activation (i.e., if the caller was
1540   // interpreted)
1541   Interpreter::_rethrow_exception_entry = __ pc();
1542   // Restore sp to interpreter_frame_last_sp even though we are going
1543   // to empty the expression stack for the exception processing.
1544   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1545   // rax: exception
1546   // rdx: return address/pc that threw exception
1547   __ restore_bcp();    // r13/rsi points to call/send
1548   __ restore_locals();
1549   LP64_ONLY(__ reinit_heapbase());  // restore r12 as heapbase.
1550   // Entry point for exceptions thrown within interpreter code
1551   Interpreter::_throw_exception_entry = __ pc();
1552   // expression stack is undefined here
1553   // rax: exception
1554   // r13/rsi: exception bcp
1555   __ verify_oop(rax);
1556   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
1557   LP64_ONLY(__ mov(c_rarg1, rax));
1558 
1559   // expression stack must be empty before entering the VM in case of
1560   // an exception
1561   __ empty_expression_stack();
1562   // find exception handler address and preserve exception oop
1563   __ call_VM(rdx,
1564              CAST_FROM_FN_PTR(address,
1565                           InterpreterRuntime::exception_handler_for_exception),
1566              rarg);
1567   // rax: exception handler entry point
1568   // rdx: preserved exception oop
1569   // r13/rsi: bcp for exception handler
1570   __ push_ptr(rdx); // push exception which is now the only value on the stack
1571   __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1572 
1573   // If the exception is not handled in the current frame the frame is
1574   // removed and the exception is rethrown (i.e. exception
1575   // continuation is _rethrow_exception).
1576   //
1577   // Note: At this point the bci is still the bxi for the instruction
1578   // which caused the exception and the expression stack is
1579   // empty. Thus, for any VM calls at this point, GC will find a legal
1580   // oop map (with empty expression stack).
1581 
1582   // In current activation
1583   // tos: exception
1584   // esi: exception bcp
1585 
1586   //
1587   // JVMTI PopFrame support
1588   //
1589 
1590   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1591   __ empty_expression_stack();
1592   // Set the popframe_processing bit in pending_popframe_condition
1593   // indicating that we are currently handling popframe, so that
1594   // call_VMs that may happen later do not trigger new popframe
1595   // handling cycles.
1596   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1597   NOT_LP64(__ get_thread(thread));
1598   __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1599   __ orl(rdx, JavaThread::popframe_processing_bit);
1600   __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1601 
1602   {
1603     // Check to see whether we are returning to a deoptimized frame.
1604     // (The PopFrame call ensures that the caller of the popped frame is
1605     // either interpreted or compiled and deoptimizes it if compiled.)
1606     // In this case, we can't call dispatch_next() after the frame is
1607     // popped, but instead must save the incoming arguments and restore
1608     // them after deoptimization has occurred.
1609     //
1610     // Note that we don't compare the return PC against the
1611     // deoptimization blob's unpack entry because of the presence of
1612     // adapter frames in C2.
1613     Label caller_not_deoptimized;
1614     Register rarg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1615     __ movptr(rarg, Address(rbp, frame::return_addr_offset * wordSize));
1616     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1617                                InterpreterRuntime::interpreter_contains), rarg);
1618     __ testl(rax, rax);
1619     __ jcc(Assembler::notZero, caller_not_deoptimized);
1620 
1621     // Compute size of arguments for saving when returning to
1622     // deoptimized caller
1623     __ get_method(rax);
1624     __ movptr(rax, Address(rax, Method::const_offset()));
1625     __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1626                                                 size_of_parameters_offset())));
1627     __ shll(rax, Interpreter::logStackElementSize);
1628     __ restore_locals();
1629     __ subptr(rlocals, rax);
1630     __ addptr(rlocals, wordSize);
1631     // Save these arguments
1632     NOT_LP64(__ get_thread(thread));
1633     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1634                                            Deoptimization::
1635                                            popframe_preserve_args),
1636                           thread, rax, rlocals);
1637 
1638     __ remove_activation(vtos, rdx,
1639                          /* throw_monitor_exception */ false,
1640                          /* install_monitor_exception */ false,
1641                          /* notify_jvmdi */ false);
1642 
1643     // Inform deoptimization that it is responsible for restoring
1644     // these arguments
1645     NOT_LP64(__ get_thread(thread));
1646     __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1647             JavaThread::popframe_force_deopt_reexecution_bit);
1648 
1649     // Continue in deoptimization handler
1650     __ jmp(rdx);
1651 
1652     __ bind(caller_not_deoptimized);
1653   }
1654 
1655   __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1656                        /* throw_monitor_exception */ false,
1657                        /* install_monitor_exception */ false,
1658                        /* notify_jvmdi */ false);
1659 
1660   // Finish with popframe handling
1661   // A previous I2C followed by a deoptimization might have moved the
1662   // outgoing arguments further up the stack. PopFrame expects the
1663   // mutations to those outgoing arguments to be preserved and other
1664   // constraints basically require this frame to look exactly as
1665   // though it had previously invoked an interpreted activation with
1666   // no space between the top of the expression stack (current
1667   // last_sp) and the top of stack. Rather than force deopt to
1668   // maintain this kind of invariant all the time we call a small
1669   // fixup routine to move the mutated arguments onto the top of our
1670   // expression stack if necessary.
1671 #ifndef _LP64
1672   __ mov(rax, rsp);
1673   __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1674   __ get_thread(thread);
1675   // PC must point into interpreter here
1676   __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1677   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1678   __ get_thread(thread);
1679 #else
1680   __ mov(c_rarg1, rsp);
1681   __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1682   // PC must point into interpreter here
1683   __ set_last_Java_frame(noreg, rbp, __ pc());
1684   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1685 #endif
1686   __ reset_last_Java_frame(thread, true);
1687 
1688   // Restore the last_sp and null it out
1689   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1690   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1691 
1692   __ restore_bcp();
1693   __ restore_locals();
1694   // The method data pointer was incremented already during
1695   // call profiling. We have to restore the mdp for the current bcp.
1696   if (ProfileInterpreter) {
1697     __ set_method_data_pointer_for_bcp();
1698   }
1699 
1700   // Clear the popframe condition flag
1701   NOT_LP64(__ get_thread(thread));
1702   __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1703           JavaThread::popframe_inactive);
1704 
1705 #if INCLUDE_JVMTI
1706   {
1707     Label L_done;
1708     const Register local0 = rlocals;
1709 
1710     __ cmpb(Address(rbcp, 0), Bytecodes::_invokestatic);
1711     __ jcc(Assembler::notEqual, L_done);
1712 
1713     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1714     // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1715 
1716     __ get_method(rdx);
1717     __ movptr(rax, Address(local0, 0));
1718     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rbcp);
1719 
1720     __ testptr(rax, rax);
1721     __ jcc(Assembler::zero, L_done);
1722 
1723     __ movptr(Address(rbx, 0), rax);
1724     __ bind(L_done);
1725   }
1726 #endif // INCLUDE_JVMTI
1727 
1728   __ dispatch_next(vtos);
1729   // end of PopFrame support
1730 
1731   Interpreter::_remove_activation_entry = __ pc();
1732 
1733   // preserve exception over this code sequence
1734   __ pop_ptr(rax);
1735   NOT_LP64(__ get_thread(thread));
1736   __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1737   // remove the activation (without doing throws on illegalMonitorExceptions)
1738   __ remove_activation(vtos, rdx, false, true, false);
1739   // restore exception
1740   NOT_LP64(__ get_thread(thread));
1741   __ get_vm_result(rax, thread);
1742 
1743   // In between activations - previous activation type unknown yet
1744   // compute continuation point - the continuation point expects the
1745   // following registers set up:
1746   //
1747   // rax: exception
1748   // rdx: return address/pc that threw exception
1749   // rsp: expression stack of caller
1750   // rbp: ebp of caller
1751   __ push(rax);                                  // save exception
1752   __ push(rdx);                                  // save return address
1753   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1754                           SharedRuntime::exception_handler_for_return_address),
1755                         thread, rdx);
1756   __ mov(rbx, rax);                              // save exception handler
1757   __ pop(rdx);                                   // restore return address
1758   __ pop(rax);                                   // restore exception
1759   // Note that an "issuing PC" is actually the next PC after the call
1760   __ jmp(rbx);                                   // jump to exception
1761                                                  // handler of caller
1762 }
1763 
1764 
1765 //
1766 // JVMTI ForceEarlyReturn support
1767 //
1768 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1769   address entry = __ pc();
1770 
1771   __ restore_bcp();
1772   __ restore_locals();
1773   __ empty_expression_stack();
1774   __ load_earlyret_value(state);  // 32 bits returns value in rdx, so don't reuse
1775 
1776   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1777   NOT_LP64(__ get_thread(thread));
1778   __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1779   Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1780 
1781   // Clear the earlyret state
1782   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1783 
1784   __ remove_activation(state, rsi,
1785                        false, /* throw_monitor_exception */
1786                        false, /* install_monitor_exception */
1787                        true); /* notify_jvmdi */
1788   __ jmp(rsi);
1789 
1790   return entry;
1791 } // end of ForceEarlyReturn support
1792 
1793 
1794 //-----------------------------------------------------------------------------
1795 // Helper for vtos entry point generation
1796 
1797 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1798                                                          address& bep,
1799                                                          address& cep,
1800                                                          address& sep,
1801                                                          address& aep,
1802                                                          address& iep,
1803                                                          address& lep,
1804                                                          address& fep,
1805                                                          address& dep,
1806                                                          address& vep) {
1807   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1808   Label L;
1809   aep = __ pc();  __ push_ptr();   __ jmp(L);
1810 #ifndef _LP64
1811   fep = __ pc(); __ push(ftos); __ jmp(L);
1812   dep = __ pc(); __ push(dtos); __ jmp(L);
1813 #else
1814   fep = __ pc();  __ push_f(xmm0); __ jmp(L);
1815   dep = __ pc();  __ push_d(xmm0); __ jmp(L);
1816 #endif // _LP64
1817   lep = __ pc();  __ push_l();     __ jmp(L);
1818   bep = cep = sep =
1819   iep = __ pc();  __ push_i();
1820   vep = __ pc();
1821   __ bind(L);
1822   generate_and_dispatch(t);
1823 }
1824 
1825 //-----------------------------------------------------------------------------
1826 
1827 // Non-product code
1828 #ifndef PRODUCT
1829 
1830 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1831   address entry = __ pc();
1832 
1833 #ifndef _LP64
1834   // prepare expression stack
1835   __ pop(rcx);          // pop return address so expression stack is 'pure'
1836   __ push(state);       // save tosca
1837 
1838   // pass tosca registers as arguments & call tracer
1839   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), rcx, rax, rdx);
1840   __ mov(rcx, rax);     // make sure return address is not destroyed by pop(state)
1841   __ pop(state);        // restore tosca
1842 
1843   // return
1844   __ jmp(rcx);
1845 #else
1846   __ push(state);
1847   __ push(c_rarg0);
1848   __ push(c_rarg1);
1849   __ push(c_rarg2);
1850   __ push(c_rarg3);
1851   __ mov(c_rarg2, rax);  // Pass itos
1852 #ifdef _WIN64
1853   __ movflt(xmm3, xmm0); // Pass ftos
1854 #endif
1855   __ call_VM(noreg,
1856              CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
1857              c_rarg1, c_rarg2, c_rarg3);
1858   __ pop(c_rarg3);
1859   __ pop(c_rarg2);
1860   __ pop(c_rarg1);
1861   __ pop(c_rarg0);
1862   __ pop(state);
1863   __ ret(0);                                   // return from result handler
1864 #endif // _LP64
1865 
1866   return entry;
1867 }
1868 
1869 void TemplateInterpreterGenerator::count_bytecode() {
1870   __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1871 }
1872 
1873 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1874   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1875 }
1876 
1877 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1878   __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1879   __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1880   __ orl(rbx,
1881          ((int) t->bytecode()) <<
1882          BytecodePairHistogram::log2_number_of_codes);
1883   __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1884   __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1885   __ incrementl(Address(rscratch1, rbx, Address::times_4));
1886 }
1887 
1888 
1889 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1890   // Call a little run-time stub to avoid blow-up for each bytecode.
1891   // The run-time runtime saves the right registers, depending on
1892   // the tosca in-state for the given template.
1893 
1894   assert(Interpreter::trace_code(t->tos_in()) != NULL,
1895          "entry must have been generated");
1896 #ifndef _LP64
1897   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1898 #else
1899   __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1900   __ andptr(rsp, -16); // align stack as required by ABI
1901   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1902   __ mov(rsp, r12); // restore sp
1903   __ reinit_heapbase();
1904 #endif // _LP64
1905 }
1906 
1907 
1908 void TemplateInterpreterGenerator::stop_interpreter_at() {
1909   Label L;
1910   __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1911            StopInterpreterAt);
1912   __ jcc(Assembler::notEqual, L);
1913   __ int3();
1914   __ bind(L);
1915 }
1916 #endif // !PRODUCT